移动机器人SLAM：理论与最新进展

5星 · 超过95%的资源需积分: 9 19 浏览量更新于2024-07-20 3 收藏 26.09MB PDF 举报

"《移动机器人同时定位与建图：原理与方法》一书由Juan-Antonio Fernández-Madrigal和José Luis Blanco Claraco两位来自西班牙马德里大学的研究者合著。本书专注于探讨移动机器人概率定位与建图理论的复杂性，并提供当前最前沿且实用的发展概述。书中深入研究了移动机器人在地理定位和导航中的核心问题，包括实时定位（Localization）和环境地图构建（Mapping）。 SLAM（Simultaneous Localization and Mapping）是现代自主导航系统的关键技术，它使机器人能够在未知环境中自主导航并持续更新其对周围环境的认识。作者通过理论分析、案例研究以及实际应用示例，展示了如何利用传感器数据融合、概率滤波器（如粒子滤波或扩展卡尔曼滤波）以及地图构建算法（如SLAM算法，如DSO、ORB-SLAM或LOAM）来解决定位与建图的挑战。书中不仅包含了基础理论的阐述，还涵盖了最新的研究成果和技术趋势，例如基于视觉的SLAM，激光雷达SLAM，以及多传感器融合的方法。此外，对于移动机器人定位误差分析、动态障碍物处理、地图更新策略等方面，都有详尽的讨论。该书适合对移动机器人技术感兴趣的工程师、研究人员以及学生，提供了宝贵的学习资源和实践指导。出版信息包括硬皮版(ISBN 978-1-4666-2104-6)，电子版(ISBN 978-1-4666-2105-3)以及打印版与永久访问权限的组合版(ISBN 978-1-4666-2106-0)。该书在图书馆学分类上属于TJ211.415，主要针对移动机器人、地理位置、定位理论等主题。此外，该书在2012年获得了国际版权，可以在大英图书馆获取目录记录。《移动机器人同时定位与建图：原理与方法》是一本全面且深入的学术著作，为理解和应用移动机器人SLAM技术提供了坚实的基础，有助于推动这一领域的进一步发展。"

Chapter 1

DOI: 10.4018/978-1-4666-2104-6.ch001

CHAPTER GUIDELINE

• You will learn:

◦ The reasons why robotic manipula-

tors have been much more successful

in practical applications than mobile

robots up to now, and the role of the

problems of localization and map-

ping in this distinction.

◦ A little on the history of mobile

robotics.

◦ How the rest of the book is organized

and how to employ it depending on

your purposes.

• Provided tools:

◦ A taxonomy for the classication

of the problems of localization and

mapping, aimed at identifying the ad-

equate approach for each situation.

• Relation to other chapters:

◦ This chapter provides an overview

of all subsequent chapters and their

organization.

1. OVERVIEW

Robotics is a complex and fascinating discipline,

not only for scientists. The idea of creating artifacts

that perform autonomously and intelligently has

appeared from time to time throughout the history

of humankind in a variety of ways, ranging from

the fearsome mythological form of the Golem of

the early Judaism (Idel, 1990) to the inoffensive

and very amusing automata of Jacques de Vau-

canson in the 18

century (Landes, 2007), until it

was definitely solidified in the last six decades of

our era. In the transition from the 19

to the 20

century, science-fiction literature was already a

Introduction

ABSTRACT

In this rst chapter of the book, the authors provide an overview of the problems of mobile robot local-

ization and mapping, including taxonomies over the axes of the representation of spatial knowledge,

the structure and dynamics of the environment, the sensory apparatus of the robot, the motor apparatus

of the robot, and the previous knowledge. They also provide a brief historical timeline and fundamental

concepts. The goal is to provide the reader with a roadmap of the problems and also of the book, in order

to allow her or him to choose the best way of approaching the text and also an appropriate understand-

ing of the main limitations of the existing methods.

Introduction

Scheinman, 1969) to parallel manipulators or the

ones used recently in space applications (Stoll,

Letschnik, Walter, Artigas, Kremer, Preusche, &

Hirzinger, 2009; Mamen, 2003), these robots have

effectively substituted repetitive and hazardous

human labor. They have been employed mainly

in the industry, where, due to their flexibility for

adaptation to variable production conditions with

a relative low cost have been one of the keys for

achieving the current development of modern

societies.

Robotic manipulators became a success in

such a short time not only for their utility, but also

for the tractability of their design (Craig, 1989;

Selig, 1992; Sandler, 1999; Kurfess, 2005), which

follows from some implicit, but important, as-

sumptions:

• Power sources do not signicantly limit

their morphology, behavior, or autonomy:

a robotic arm can assume an unlimited sup-

ply of energy at every moment, disregard-

ing unavoidable power shortages or other

sporadic incidents external to the robotic

system.

• Their working environment is basically

invariant, of very limited extension, and

known in advanced by the robot designers.

Therefore, a nearly complete planning of

the activities of the arm can be carried out

before the robot moves into action.

• Not only the environment is static and

small, but also the physical properties of

the objects that the robotic arm has to ma-

nipulate are known in detail before opera-

tion; that is necessary in order to include

the robot as part of an efcient production

system. In addition, there are restrictions

in the possible spatial location of those ob-

jects with respect to the robot.

• The state space of the arm—the angular

position of all its joints plus their cor-

responding dynamics—is restricted to a

bounded volume.

In contrast, designing a robot that moves on

its own throughout the world while executing

some orders (see Figure 2) is a completely differ-

ent story (Cox, Wilfong, & Lozano-Pérez, 1990;

Siegwart & Nourbakhsh, 2004). Problems multi-

Figure 2. Some of the mobile robots developed and used in the machine perception and intelligent

robotics group (MAPIR) of the system engineering and automation Dpt. of the University of Málaga,

where the authors have implemented a number of systems for localization and mapping, among other

robotic problems.

Introduction

ply in number and complexity, quickly becoming

mathematically intractable. To begin with, we

have the fundamental issue of power supply. A

mobile robot, just like fixed manipulators, should

be able of performing autonomously (i.e., without

external intervention) during long periods of time

operation; but these robots has to carry their own

power source on board, and any mobile platform

has restrictions regarding the maximum weight

of its payload. Closing the loop, power can only

be increased at the expense of larger and heavier

power supplies. This issue is analogous to the

one that limits the size and autonomy of space

ships—which led to the invention of multistage

rockets—or, for the matter, the one confronted by

all kinds of automobiles. It is not the only chicken

or the egg dilemma that we will find within mobile

robotics. If you have thought of the robot going

periodically and autonomously to a recharging

station (i.e., finding it, approaching it, using it),

please read the following issues.

A mobile robot must, by definition, change its

location in order to perform it tasks, which tears

down most of the other benefits enjoyed by ro-

botic manipulators. Free motion makes the current

environment of the robot vary, even largely, from

one moment to another. Occasionally, engineers

can build a model of the whole workspace, but in

most practical applications, it is common having

to face at least partially unknown regions. It is

difficult to plan any complex sequence of actions

with limited knowledge of their effects in all the

particular scenarios that the robot may encounter:

possibilities explode exponentially with time, and

thus planning must be stopped regularly to per-

ceive, analyze, and correct the results. Sensors

become crucial devices for mobile robots pre-

cisely for that reason. The process of transforming

sensory data into useful information, i.e., percep-

tion, arises then with all its complexity.

Motion also leads to the possibility of operating

in spaces that are potentially unlimited in size (that

is one of the main advantages of mobile robots,

after all!). The issues that can arise from this are

obvious for a programmable machine that has

finite memory and computational capabilities—

which, again, can only be increased at the expense

of more size and weight, feeding again the beast

of the power supply. Therefore, approximations

may have to be done even for problems otherwise

solvable. Furthermore, aside from computational

limitations, we find the issue of merging all the

information gathered at different locations of a

large environment in order to get a fairly complete

representation of it, which is a prerequisite for

planning future tasks.

Finally, the objects, people and other robots or

machines that a mobile robot is to interact with

during its performance can also be considered as

part of its environment, and as such they have a

relevant influence in not having a complete, static,

small model of the physical world. Unexpected

obstacles with arbitrary shapes and unknown

dynamics are one of the common problems that a

real mobile robot will have to face without much

previous knowledge. If those obstacles are intel-

ligent beings, understanding and predicting their

behavior becomes especially challenging.

In summary, the effects of including the capa-

bility of free motion in a robot lead to an important

increase in dimensionality, size, and complexity

of nearly all the issues of robotics at very different

levels, from mechanics to electronics to computer

science to artificial intelligence. Most of them

have no satisfactory solution yet—albeit many

proposals—, and that is the reason why today we

do not see mobile robots everywhere in the real

world, especially operating reliably during long

periods of time. It is also the reason why mobile

robotics remains such a fascinating discipline. It

will probably be for quite a while.

In spite of all these difficulties, there is one

basic problem of mobile robotics that has reached

a reasonable maturity in the last decades, up to the

point of being considered satisfactorily solved in

many practical situations. It has to do with mobile

robot navigation, that is, taking the robot from

one point to another. Maybe the reader who is

Introduction

unfamiliar with mobile robotics is surprised by

the apparent triviality of the problem. Six decades

just to make the robot go well from one place to

another, the most basic of the tasks of a mobile

robot? For now, we hope that the previous ratio-

nale, besides the fact that a book like this one is

needed just to introduce two of the main basic

aspects of navigation, may convince him or her

that solving this problem is actually a big deal.

Nevertheless, robotics is a highly active area of

research, and, apart from these problems, many

other advances are succeeding concerning other

issues (Siciliano & Khatib, 2005).

Let us focus on navigation then. It is obvious

that a mobile robot must be able to navigate, that

is, to change its location autonomously, since

it performs operations that are distributed over

space. It is also obvious, and much less striking

at first sight, that in order to move to a target

position the robot needs to figure out the spatial

relation existing between that target and its cur-

rent location. That target-robot relation is of the

uttermost importance: it is needed for deciding

actions, for example the motion of the robot actua-

tors (Krzysztof, 2006) that reduce the distance to

the target. Being able to estimate that relation is

a basic, unavoidable step; it will be required for

any navigation process.

The target-robot relation can be expressed in a

variety of forms, mainly quantitative (geometrical,

e.g. two points in a Euclidean coordinate reference

system) or qualitative (a set of logical relations

between places, such as “to the right of” or “far-

ther than”), even in a mixture of both. How it is

represented defines in turn the methods the robot

may use to navigate. Scientific literature on mobile

robot navigation algorithms has been quite prolific

since the 1980s (Lozano-Pérez & Wesley, 1979;

Thorpe, 1984; Crowley, 1985; Zheng & Zheng,

1994; Borenstein, Everett, & Feng, 1996; Siegwart

& Nourbakhsh, 2004; Minguez & Montano, 2008;

Blanco, González, & Fernández-Madrigal, 2008),

but we are especially interested here just in the

mentioned aspect: the relation existing between

the target and the robot, and how it is acquired

in order to navigate, rather than in the manner of

performing that navigation.

Estimating the target-robot relation in a pre-

viously unmodeled environment can become a

tough problem, which is clear when we realize

that it involves handling sensory data reflecting

unknown elements of the environment, as seen

from unknown positions. It is worth mentioning

that a robot can navigate without any representa-

tion—implicit or explicit—of the environment, but

in practice such an approach limits the navigation

autonomy to only small distances: sooner or later,

the robot finds difficulties locating the target or

itself since errors accumulate if it tracks its current

position solely through odometry (Borenstein,

1998). Thus, at some point mobile robots have

to use information from their environment to

navigate.

And that is how we arrive at the main issues

we are interested in: using known environmental

information, that is, relations existing among

environment elements, for estimating the actual

position of the robot is the problem called local-

ization in mobile robotics—less popular in the

specialized literature is the term positioning—to

which the second part of this book is entirely

dedicated; it is one of the aspects of navigation

that we labeled above as “satisfactorily solved in

practical situations.” Today, most mobile robots

can localize themselves and their targets automati-

cally and accurately as long as enough environment

and sensory information is available, although

the mathematical tools used in the problem are

far from simple—that, in turn, is the reason for

the size of the first part of the book: we need to

provide the reader with those foundations.

On the other hand, estimating the unknown

spatial relations that exist between environment

elements in order to build an internal representa-

tion suitable for robot navigation is the problem

called mapping. The third and last part of the book

addresses this problem, which can also be con-

sidered as satisfactorily solved in some practical

剩余497页未读，继续阅读

BiscuitET

粉丝: 0
资源: 3

移动机器人SLAM：理论与最新进展

Mapping of stream microhabitats with high spatial

Simultaneous Localization and Mapping for Mobile Robots

Multi-session Visual Simultaneous Localisation and Mapping（移动机器人slam）

如何在ROS（Robot Operating System）环境中有效地切换map_server使用的SLAM（Simultaneous Localization And Mapping）地图？

ORB-SLAM全称

slam分层强化学习算法

ORB-SLAM2全称

yolo5 orbslam2 UNIX

UBUNTU SLAM 算法案例

yolo orbslam

最新资源